Cardano founder Charles Hoskinson has turned his attention to artificial intelligence hardware, and argued that Apple could dramatically lower the cost of running large AI models locally compared with Nvidia-based systems.
In a recent post on X, Hoskinson said he was surprised by how far Apple has progressed with local large language models. His focus was on the potential of Apple’s next-generation Mac Studio hardware and the amount of unified memory that could be combined across several machines.
“My mind is blown by how far ahead Apple is with local LLMs,” Hoskinson wrote.
He estimated that a Mac Studio equipped with an M5 Ultra and 512 GB of memory could cost roughly $15,000 to $20,000. In his scenario, connecting six machines through Thunderbolt 5 would provide more than 3 TB of combined memory alongside substantial CPU and GPU resources.
Hoskinson Sees a Big Cost Gap With Nvidia
Hoskinson argued that such a setup could potentially run a frontier AI model containing more than two trillion parameters locally for approximately $100,000.
He contrasted that with Nvidia-based infrastructure, claiming comparable capability could cost closer to $500,000.
My mind is blown by how far ahead Apple is with local LLMs. The new Mac Studio M5 Ultra will have a 512 GB model for around 15-20k. You can chain 6 of them together using thunderbolt 5 and this have over 3 TB of ram and a massive amount of CPU and GPU cores. This means you can run a frontier model with over 2 trillion parameters locally for around 100kish. Nivida is around 500k for the same capability.
— Charles Hoskinson (@IOHK_Charles) October 18, 2026
That would represent an approximately 80% cost difference if the comparison holds under real-world workloads.
The important caveat is that these are Hoskinson’s estimates, not independently benchmarked performance figures. Running a model with trillions of parameters also depends on factors beyond total memory, including model precision, bandwidth, interconnect performance, inference software and how efficiently workloads can be distributed across machines.
Still, his broader point is pretty significant.
Read also: “Black Monday” Fears Explode, But Bitcoin Price Tops $83K
If high-end local AI becomes considerably cheaper, developers and companies could run increasingly capable models without relying entirely on centralized cloud infrastructure.
For Hoskinson, Apple’s architecture may be pushing that economics in a direction that could challenge Nvidia’s dominance in certain local AI workloads.
For more crypto news and price predictions, click here.
Subscribe to our YouTube channel for daily crypto updates, market insights, and expert analysis.
